A ranking-based problem transformation method for weakly supervised multi-label learning

Jiaxuan Li, Xiaoyan Zhu, Weichu Zhang, Jiayin Wang

Abstract: Highlights•A problem transformation methodology is proposed for multi-label learning (MLL).•The novel method is proposed named EPR (Ensemble of Pairwise Ranking learners).•EPR is compatible with weakly supervised MLL via pairwise label correlation learning.•EPR transforms weakly supervised multi-label learning into pairwise ranking patterns.•EPR is available for partial MLL, noisy MLL, and MLL with missing labels.•A heuristic ensemble pruning mechanism is designed balancing accuracy and efficiency.